AI Paper Atlas

Research map / Large language models / Language, safety and interpretability

Detecting and Filtering Unsafe Training Data via Data Attribution with Denoised Representation

Yijun Pan, Taiwei Shi, Jieyu Zhao, Jiaqi W. Ma

ICML 2026 · Large language models › Language, safety and interpretability · approach: Data and hallucination detection · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
apply DRA denoising to GradSimGradSim (Llama-3-8B-Instruct) · XSTest-Response detectionAUPRC (%): 1.7 -> 81.5Improved
apply DRA denoising to RepSimRepSim (Gemma-2-9B-it) · ToxicChat detectionAUPRC (%): 51.6 -> 98.2Improved
apply DRA to GradSim before filtering and retrainingGradSim (Gemma-2-9B-it) · XSTest-Response retrain ASRASR: 77.0% -> 3.0%Improved
apply DRA to GradSim before filtering and retrainingGradSim (Llama-3-8B-Instruct) · XSTest-Response retrain ASRASR: 68.0% -> 43.0%Improved
apply DRA to RepSim before filtering and retrainingRepSim (Llama-3-8B-Instruct) · XSTest-Response retrain ASRASR: 63.0% -> 60.0%Improved
apply DRA to RepSim before filtering and retrainingRepSim (Gemma-2-9B-it) · ToxicChat retrain ASRASR: 29.0% -> 30.7%Hurt
apply DRA to GradSimGradSim (Llama-3-8B-Instruct) · Bias in Bios detectionAUPRC: 0.417 -> 0.550Improved
apply DRA to RepSimRepSim (Llama-3-8B-Instruct) · Bias in Bios detectionAUPRC: 0.333 -> 0.748Improved
apply DRA to GradSim before filtering top 500 and retrainingGradSim (Llama-3-8B-Instruct) · Bias in Bios retrainTPR gender gap: 0.650 -> 0.246Improved
apply DRA to RepSim before filtering top 500 and retrainingRepSim (Gemma-2-9B-it) · Bias in Bios retrainTPR gender gap: 0.536 -> 0.346Improved
use Wildguard moderation classifier for filteringWildguard · Bias in BiosAUPRC: 0.059No clear effect
increase number of target examples for DRA (ablation)DRA · jailbreak detectionAUPRCImproved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy